[ovirt-users] Re: CEPH - Opinions and ROI

Yes, we manage a number of Distributed Storage systems including MooseFS, Ceph, DRBD and of course Gluster (since 3.3). Each has a specific use. For small customer-specific VM host clusters, which is the majority of what we do, Gluster is by far the safest and easiest to deploy/understand for the more junior members of the team. We have never lost a VM image on Gluster, which can't be said about the others (including CEPH but that disaster was years ago and somewhat self-inflicted). The point is that it hard to shoot yourself in the foot with Gluster. The newer innovations on Gluster such as sharding and the arbiter node have allowed it be competitive on the performance/hassle factor. Our Ceph cluster is on one of the few larger host installations we have and is mostly handled by a more senior tech who has lots of experience with it. He clearly loves it and doesn't understand why we aren't fans but it just seems to be overkill for the typical 3 host VM cluster. The rest of us worry about him getting hit by a bus. For the record I really like MooseFS, but not for live VMs, we use it for archiving and it is the easiest to maintain as long as you are paranoid with the "master" server which provides the metadata index for the chunkserver nodes. My hope for Gluster is that it is able to continue to improve with some of the new ideas such as the thin-arbiter and keep that performance/hassle ratio high. My worry is that IBM/Redhat makes more money on Ceph consulting, than Gluster and thus contributes to the idea that Gluster is a deprecated technology. On 10/1/2020 7:53 AM, Strahil Nikolov via Users wrote: > CEPH requires at least 4 nodes to be "good". > I know that Gluster is not the "favourite child" for most vendors, yet it is still optimal for HCI. > > You can check https://www.ovirt.org/develop/release-management/features/storage/c... for cinder integration. > > Best Regards, > Strahil Nikolov > > >